A PERSPECTIVE-BASED COMPLEXITY FRAMEWORK AND METRICS FOR PREDICTING THE ANALYZABILITY OF UNIFIED MODELING LANGUAGE BEHAVIORAL DIAGRAMS

Abstract

Software designers are increasingly adopting Unified Modeling Language (UML) behavioral diagrams to communicate a system's dynamic behavior to other software development team members. UML behavioral diagrams have inherent complexity which gets more complex whenever the diagrams evolve. This leads to complex systems that are not easy to analyze. Although several metric measurement frameworks and measures have been proposed in the literature, they cannot be applied to analyze the complexity of these diagrams due to their limitations. This is because behavioral diagrams have unique features such as elements, control flow, and interaction. The main objective of this study was to develop a perspective-based metrics framework for predicting the analyzability of UML behavioral diagrams. Specifically, the study sought to identify relevant complexity perspectives, design a metrics theoretical validation framework, define perspective-based metrics, develop a metrics automation tool, and empirically evaluate the predictive capability of the metrics. To achieve these objectives, a pragmatism philosophy was chosen. Therefore, a mixed research methodology was adopted to solve the problem. A qualitative approach was used to identify the perspective-based information for the definition and theoretical validation of metrics. Further, a quantitative approach was used to investigate the usefulness of the metrics in predicting the analyzability of the metrics. A metrics tool named Behavioral Complexity Metrics Tool (BCMT) was developed to automate the computation of metric values from the diagrams. The BCMT tool and the structured questionnaire were employed for data collection. The snowballing and systematic sampling techniques were employed to select 11 experts to validate the perspective-based complexity framework and 58 subjects for validation of complexity metrics. The data collected was analyzed through descriptive statistics (frequencies, mean, range, maximum, minimum and standard deviation) and inferential statistics (correlation, and regression analysis). Results reveal that the proposed Perspective￾based Complexity Framework (PBCF) can be relied upon as a complexity taxonomy for UML behavioral diagrams. Also, findings indicate that the Perspective-based Metrics Validation Framework (PBMVF) provides a comprehensive and systematic approach of evaluating behavioral metrics. Empirical validations were conducted, and results indicate that the metrics are strongly correlated with analyzability and time to analyze diagrams, and the regression model showed that the metrics are predictors of analyzability. These results will particularly aid the software designers, to guide in decision making on how to design systems that are easy to analyze hence ensuring the quality of these diagrams. It is therefore recommended that the framework and metrics be adopted in software design practices and tools to support the development of more analyzable systems. In conclusion, the proposed metrics and framework are reliable, practical, and contribute significantly to improving the analyzability and overall quality of UML behavioral diagrams.